Relation Extraction Based on Multidimensional Semantic Mapping

Abstract

Relation extraction aims to identify relation types between entities from texts.In the field of relation extraction,most of existing methods use deep learning methods,but they do not have in-depth discussion of word vectors in the input layer.To further exploit word vectors,this paper proposes a relation extraction method based on multi-dimensional semantic mapping.The core idea of the method is to reduce dimensionality of text feature matrix before the word vector enters the input layer.Experimental results show that the proposed method not only can reduce dimensionality effectively,but also can represent the semantic information of the same sentence in different dimensions,with its F1 of 75.3% and 88.9% on the Chinese Literature Text and SemEval-2010 Task8 datasets,respectively.

Keywords

• relation extraction|neural network|multidimensional mapping|semantic information